When someone talks about “AI”, today, they are usually referring to one particular type of AI: multi-layer neural nets trained on big data to recognize patterns. These so-called “deep learning” algorithms are great at learning more or less the same sort of stimulus/response functionality that our right brain hemispheres carry out – what Daniel Kahneman calls “thinking fast”. Almost all of today’s AI’s have little or no left brain function – logical, causal, “thinking slowly”. Homo sapiens pays a huge price for having an over-sized bicameral brain (high birthing pain and risk) but upon reflection it’s worth it — in particular, you or I couldn’t perform such reflection without it! Our ability to generate alternative possibilities, to rationally construct and consider pro- and con- arguments for and against each proposition, is clearly worth it. At least for decisions that are important to us.
There are three qualitatively different types or levels of AI understanding:
- Machine Learning
- Limited Logics
- Higher Order Logics
Cycorp’s founder and CEO, Dr. Doug Lenat, entertainingly explores the similarities, differences, strengths and limitations of each in an article published in his regular Forbes column titled What AI Can Learn from Romeo and Juliet.